Objective:To evaluate the prediction value of constructing Logistic regression model based on the preoperative detection indexes in the pathological types of periampullary carcinoma.Methods:256 patients with periampullary carcinoma who underwent radical pancreaticoduodenectomy in the First Affiliated Hospital of Xi'an Jiaotong University from January 2015 to December 2018 were recruited in this clinical trial. Among them, 159 patients were male and 97 female, aged from 39 to 88 years, with a median age of 64 years. The informed consents of all patients were obtained and the local ethical committee approval was received. Preoperative data and postoperative pathological types of all patients were collected, and Logistic regression model, decision tree model and random forest model were established to predict the pathological types of periampullary carcinoma. 192 cases were randomly selected into the training set and 64 cases into the test set. The predictive value of this model was assessed by the area under ROC curve (AUC).Results:Among all patients, 99 patients were classified as the intestinal type and 157 cases of pancreaticobiliary type. Based on the tumor diameter and Hb variables, a multivariate Logistic regression model was constructed, and the Akaike information criterion (AIC) value reached the lowest 24. According to the results of AdaBoost algorithm, CEA, CA125, ALB, tumor diameter and CA19-9 were important factors affecting the decision tree. The corresponding decision tree model and random forest model were obtained by setting random seeds as 91 and 62, respectively. In the test set, the accuracy of Logistic regression, decision tree and random forest models was 0.810, 0.766 and 0.766, the sensitivity was 0.848, 0.775 and 0.956, the specificity was 0.706, 0.750 and 0.316, and the AUC was 0.764, 0.762 and 0.698, respectively.Conclusions:Logistic regression model based on tumor diameter and Hb can accurately identify the pathological types of periampullary carcinoma before surgery, and provide guidance for the preoperative adjuvant chemotherapy.
万方数据
